summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--usr/src/cmd/sgs/include/libld.h3
-rw-r--r--usr/src/cmd/sgs/libld/common/files.c4
-rw-r--r--usr/src/cmd/truss/codes.c5
-rw-r--r--usr/src/man/man2/open.212
-rw-r--r--usr/src/uts/common/io/mac/mac_sched.c310
-rw-r--r--usr/src/uts/common/sys/fcntl.h3
-rw-r--r--usr/src/uts/common/sys/file.h4
-rw-r--r--usr/src/uts/common/syscall/open.c9
8 files changed, 159 insertions, 191 deletions
diff --git a/usr/src/cmd/sgs/include/libld.h b/usr/src/cmd/sgs/include/libld.h
index d0d4776220..ddb0fb2506 100644
--- a/usr/src/cmd/sgs/include/libld.h
+++ b/usr/src/cmd/sgs/include/libld.h
@@ -891,6 +891,7 @@ struct ifl_desc { /* input file descriptor */
/* symbol capabilities */
#define FLG_IF_DEFERRED 0x00080000 /* dependency is deferred */
#define FLG_IF_RTLDINF 0x00100000 /* dependency has DT_SUNW_RTLTINF set */
+#define FLG_IF_GROUPS 0x00200000 /* input file has groups to process */
/*
* Symbol states that require the generation of a DT_POSFLAG_1 .dynamic entry.
@@ -934,7 +935,7 @@ struct is_desc { /* input section descriptor */
#define FLG_IS_EXTERNAL 0x0040 /* isp from a user file */
#define FLG_IS_INSTRMRG 0x0080 /* Usable SHF_MERGE|SHF_STRINGS sec */
#define FLG_IS_GNSTRMRG 0x0100 /* Generated mergeable string section */
-#define FLG_IS_GROUPS 0x0200 /* section has groups to process */
+
#define FLG_IS_PLACE 0x0400 /* section requires to be placed */
#define FLG_IS_COMDAT 0x0800 /* section is COMDAT */
#define FLG_IS_EHFRAME 0x1000 /* section is .eh_frame */
diff --git a/usr/src/cmd/sgs/libld/common/files.c b/usr/src/cmd/sgs/libld/common/files.c
index c1e66ecfea..99e6e89f9d 100644
--- a/usr/src/cmd/sgs/libld/common/files.c
+++ b/usr/src/cmd/sgs/libld/common/files.c
@@ -2343,7 +2343,7 @@ process_group(const char *name, Ifl_desc *ifl, Shdr *shdr, Elf_Scn *scn,
* Indicate that this input file has groups to process. Groups are
* processed after all input sections have been processed.
*/
- ifl->ifl_flags |= FLG_IS_GROUPS;
+ ifl->ifl_flags |= FLG_IF_GROUPS;
return (1);
}
@@ -2807,7 +2807,7 @@ process_elf(Ifl_desc *ifl, Elf *elf, Ofl_desc *ofl)
* of placement. In addition, COMDAT names may require transformation
* to indicate different output section placement.
*/
- if (ifl->ifl_flags & FLG_IS_GROUPS) {
+ if (ifl->ifl_flags & FLG_IF_GROUPS) {
for (ndx = 1; ndx < ifl->ifl_shnum; ndx++) {
Is_desc *isp;
diff --git a/usr/src/cmd/truss/codes.c b/usr/src/cmd/truss/codes.c
index 4162c8b661..dcab90e5fa 100644
--- a/usr/src/cmd/truss/codes.c
+++ b/usr/src/cmd/truss/codes.c
@@ -24,6 +24,7 @@
* Copyright (c) 2012 by Delphix. All rights reserved.
* Copyright 2011 Nexenta Systems, Inc. All rights reserved.
* Copyright (c) 2012, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved.
*/
/* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
@@ -1918,7 +1919,7 @@ pathconfname(int code)
#define ALL_O_FLAGS \
(O_NDELAY|O_APPEND|O_SYNC|O_DSYNC|O_NONBLOCK|O_CREAT|O_TRUNC\
|O_EXCL|O_NOCTTY|O_LARGEFILE|O_RSYNC|O_XATTR|O_NOFOLLOW|O_NOLINKS\
- |FXATTRDIROPEN)
+ |O_CLOEXEC|FXATTRDIROPEN)
const char *
openarg(private_t *pri, int arg)
@@ -1976,6 +1977,8 @@ openarg(private_t *pri, int arg)
(void) strlcat(str, "|O_NOFOLLOW", sizeof (pri->code_buf));
if (arg & O_NOLINKS)
(void) strlcat(str, "|O_NOLINKS", sizeof (pri->code_buf));
+ if (arg & O_CLOEXEC)
+ (void) strlcat(str, "|O_CLOEXEC", sizeof (pri->code_buf));
if (arg & FXATTRDIROPEN)
(void) strlcat(str, "|FXATTRDIROPEN", sizeof (pri->code_buf));
diff --git a/usr/src/man/man2/open.2 b/usr/src/man/man2/open.2
index 30ae089ca7..522ac12e1e 100644
--- a/usr/src/man/man2/open.2
+++ b/usr/src/man/man2/open.2
@@ -2,6 +2,7 @@
.\" Copyright (c) 2008, Sun Microsystems, Inc. All Rights Reserved.
.\" Copyright 1989 AT&T
.\" Portions Copyright (c) 1992, X/Open Company Limited. All Rights Reserved.
+.\" Portions Copyright (c) 2013, OmniTI Computer Consulting, Inc. All Rights Reserved.
.\" Sun Microsystems, Inc. gratefully acknowledges The Open Group for permission to reproduce portions of its copyrighted documentation. Original documentation from The Open Group can be obtained online at
.\" http://www.opengroup.org/bookstore/.
.\" The Institute of Electrical and Electronics Engineers and The Open Group, have given us permission to reprint portions of their documentation. In the following statement, the phrase "this text" refers to portions of the system documentation. Portions of this text are reprinted and reproduced in electronic form in the Sun OS Reference Manual, from IEEE Std 1003.1, 2004 Edition, Standard for Information Technology -- Portable Operating System Interface (POSIX), The Open Group Base Specifications Issue 6, Copyright (C) 2001-2004 by the Institute of Electrical and Electronics Engineers, Inc and The Open Group. In the event of any discrepancy between these versions and the original IEEE and The Open Group Standard, the original IEEE and The Open Group Standard is the referee document. The original Standard can be obtained online at http://www.opengroup.org/unix/online.html.
@@ -214,6 +215,17 @@ sets \fBerrno\fR to \fBEMLINK\fR.
.sp
.ne 2
.na
+\fB\fBO_CLOEXEC\fR\fR
+.ad
+.sp .6
+.RS 4n
+If set, the file descriptor returned will be closed prior to any future
+\fBexec()\fR calls.
+.RE
+
+.sp
+.ne 2
+.na
\fB\fBO_NONBLOCK\fR or \fBO_NDELAY\fR\fR
.ad
.sp .6
diff --git a/usr/src/uts/common/io/mac/mac_sched.c b/usr/src/uts/common/io/mac/mac_sched.c
index 06a5ac8cbf..bb6fc84294 100644
--- a/usr/src/uts/common/io/mac/mac_sched.c
+++ b/usr/src/uts/common/io/mac/mac_sched.c
@@ -21,9 +21,8 @@
/*
* Copyright 2010 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
- */
-/*
* Copyright 2011 Joyent, Inc. All rights reserved.
+ * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
*/
#include <sys/types.h>
@@ -530,10 +529,10 @@ enum pkt_type {
/*
* In general we do port based hashing to spread traffic over different
- * softrings. The below tunables allows to override that behavior. Setting it
- * to B_TRUE allows to do a fanout based on src ipv6/ipv4 address. This behavior
- * is also applicable to ipv6 packets carrying multiple optional headers
- * and other uncommon packet types.
+ * softrings. The below tunables allow to override that behavior. Setting one
+ * (depending on IPv6 or IPv4) to B_TRUE allows a fanout based on src
+ * IPv6 or IPv4 address. This behavior is also applicable to IPv6 packets
+ * carrying multiple optional headers and other uncommon packet types.
*/
boolean_t mac_src_ipv6_fanout = B_FALSE;
boolean_t mac_src_ipv4_fanout = B_FALSE;
@@ -761,231 +760,168 @@ mac_rx_srs_proto_fanout(mac_soft_ring_set_t *mac_srs, mblk_t *head)
}
}
-int fanout_unalligned = 0;
+int fanout_unaligned = 0;
/*
* mac_rx_srs_long_fanout
*
- * The fanout routine for IPv6 (and IPv4 when VLANs are in use).
+ * The fanout routine for VLANs, and for anything else that isn't performing
+ * explicit dls bypass. Returns -1 on an error (drop the packet due to a
+ * malformed packet), 0 on success, with values written in *indx and *type.
*/
static int
mac_rx_srs_long_fanout(mac_soft_ring_set_t *mac_srs, mblk_t *mp,
uint32_t sap, size_t hdrsize, enum pkt_type *type, uint_t *indx)
{
ip6_t *ip6h;
- struct ip *ip4h;
+ ipha_t *ipha;
uint8_t *whereptr;
uint_t hash;
uint16_t remlen;
uint8_t nexthdr;
uint16_t hdr_len;
+ uint32_t src_val;
+ boolean_t modifiable = B_TRUE;
+ boolean_t v6;
- if (sap == ETHERTYPE_IPV6) {
- boolean_t modifiable = B_TRUE;
+ ASSERT(MBLKL(mp) >= hdrsize);
- ASSERT(MBLKL(mp) >= hdrsize);
+ if (sap == ETHERTYPE_IPV6) {
+ v6 = B_TRUE;
+ hdr_len = IPV6_HDR_LEN;
+ } else if (sap == ETHERTYPE_IP) {
+ v6 = B_FALSE;
+ hdr_len = IP_SIMPLE_HDR_LENGTH;
+ } else {
+ *indx = 0;
+ *type = OTH;
+ return (0);
+ }
- ip6h = (ip6_t *)(mp->b_rptr + hdrsize);
- if ((unsigned char *)ip6h == mp->b_wptr) {
- /*
- * The first mblk_t only includes the mac header.
- * Note that it is safe to change the mp pointer here,
- * as the subsequent operation does not assume mp
- * points to the start of the mac header.
- */
- mp = mp->b_cont;
+ ip6h = (ip6_t *)(mp->b_rptr + hdrsize);
+ ipha = (ipha_t *)ip6h;
- /*
- * Make sure ip6h holds the full ip6_t structure.
- */
- if (mp == NULL)
- return (-1);
+ if ((uint8_t *)ip6h == mp->b_wptr) {
+ /*
+ * The first mblk_t only includes the mac header.
+ * Note that it is safe to change the mp pointer here,
+ * as the subsequent operation does not assume mp
+ * points to the start of the mac header.
+ */
+ mp = mp->b_cont;
- if (MBLKL(mp) < IPV6_HDR_LEN) {
- modifiable = (DB_REF(mp) == 1);
+ /*
+ * Make sure the IP header points to an entire one.
+ */
+ if (mp == NULL)
+ return (-1);
- if (modifiable &&
- !pullupmsg(mp, IPV6_HDR_LEN)) {
- return (-1);
- }
- }
+ if (MBLKL(mp) < hdr_len) {
+ modifiable = (DB_REF(mp) == 1);
- ip6h = (ip6_t *)mp->b_rptr;
+ if (modifiable && !pullupmsg(mp, hdr_len))
+ return (-1);
}
- if (!modifiable || !(OK_32PTR((char *)ip6h)) ||
- ((unsigned char *)ip6h + IPV6_HDR_LEN > mp->b_wptr)) {
- /*
- * If either ip6h is not alligned, or ip6h does not
- * hold the complete ip6_t structure (a pullupmsg()
- * is not an option since it would result in an
- * unalligned ip6h), fanout to the default ring. Note
- * that this may cause packets reordering.
- */
- *indx = 0;
- *type = OTH;
- fanout_unalligned++;
- return (0);
- }
+ ip6h = (ip6_t *)mp->b_rptr;
+ ipha = (ipha_t *)ip6h;
+ }
+ if (!modifiable || !(OK_32PTR((char *)ip6h)) ||
+ ((uint8_t *)ip6h + hdr_len > mp->b_wptr)) {
+ /*
+ * If either the IP header is not aligned, or it does not hold
+ * the complete simple structure (a pullupmsg() is not an
+ * option since it would result in an unaligned IP header),
+ * fanout to the default ring.
+ *
+ * Note that this may cause packet reordering.
+ */
+ *indx = 0;
+ *type = OTH;
+ fanout_unaligned++;
+ return (0);
+ }
+
+ /*
+ * Extract next-header, full header length, and source-hash value
+ * using v4/v6 specific fields.
+ */
+ if (v6) {
remlen = ntohs(ip6h->ip6_plen);
nexthdr = ip6h->ip6_nxt;
-
- if (remlen < MIN_EHDR_LEN)
- return (-1);
+ src_val = V4_PART_OF_V6(ip6h->ip6_src);
/*
* Do src based fanout if below tunable is set to B_TRUE or
* when mac_ip_hdr_length_v6() fails because of malformed
- * packets or because mblk's need to be concatenated using
+ * packets or because mblks need to be concatenated using
* pullupmsg().
*/
if (mac_src_ipv6_fanout || !mac_ip_hdr_length_v6(ip6h,
mp->b_wptr, &hdr_len, &nexthdr, NULL)) {
- goto ipv6_src_based_fanout;
- }
- whereptr = (uint8_t *)ip6h + hdr_len;
-
- /* If the transport is one of below, we do port based fanout */
- switch (nexthdr) {
- case IPPROTO_TCP:
- case IPPROTO_UDP:
- case IPPROTO_SCTP:
- case IPPROTO_ESP:
- /*
- * If the ports in the transport header is not part of
- * the mblk, do src_based_fanout, instead of calling
- * pullupmsg().
- */
- if (mp->b_cont != NULL &&
- whereptr + PORTS_SIZE > mp->b_wptr) {
- goto ipv6_src_based_fanout;
- }
- break;
- default:
- break;
- }
-
- switch (nexthdr) {
- case IPPROTO_TCP:
- hash = HASH_ADDR(V4_PART_OF_V6(ip6h->ip6_src),
- *(uint32_t *)whereptr);
- *indx = COMPUTE_INDEX(hash,
- mac_srs->srs_tcp_ring_count);
- *type = OTH;
- break;
-
- case IPPROTO_UDP:
- case IPPROTO_SCTP:
- case IPPROTO_ESP:
- if (mac_fanout_type == MAC_FANOUT_DEFAULT) {
- hash = HASH_ADDR(V4_PART_OF_V6(ip6h->ip6_src),
- *(uint32_t *)whereptr);
- *indx = COMPUTE_INDEX(hash,
- mac_srs->srs_udp_ring_count);
- } else {
- *indx = mac_srs->srs_ind %
- mac_srs->srs_udp_ring_count;
- mac_srs->srs_ind++;
- }
- *type = OTH;
- break;
-
- /* For all other protocol, do source based fanout */
- default:
- goto ipv6_src_based_fanout;
+ goto src_based_fanout;
}
- } else if (sap == ETHERTYPE_IP) {
- boolean_t modifiable = B_TRUE;
-
- ASSERT(MBLKL(mp) >= hdrsize);
-
- ip4h = (struct ip *)(mp->b_rptr + hdrsize);
-
- if ((unsigned char *)ip4h == mp->b_wptr) {
- /*
- * The first mblk_t only includes the mac header.
- * Note that it is safe to change the mp pointer here,
- * as the subsequent operation does not assume mp
- * points to the start of the mac header.
- */
- mp = mp->b_cont;
-
- /*
- * Make sure ip4h holds the full base ip structure
- * up through the destination address. It might not
- * hold any of the options though.
- */
- if (mp == NULL)
- return (-1);
-
- if (MBLKL(mp) < IP_SIMPLE_HDR_LENGTH) {
- modifiable = (DB_REF(mp) == 1);
-
- if (modifiable &&
- !pullupmsg(mp, IP_SIMPLE_HDR_LENGTH))
- return (-1);
- }
-
- ip4h = (struct ip *)mp->b_rptr;
- }
-
- if (!modifiable || !(OK_32PTR((char *)ip4h))) {
- /*
- * If ip4h is not aligned fanout to the default ring.
- * Note that this may cause packets reordering.
- */
- *indx = 0;
- *type = OTH;
- fanout_unalligned++;
- return (0);
+ } else {
+ hdr_len = IPH_HDR_LENGTH(ipha);
+ remlen = ntohs(ipha->ipha_length) - hdr_len;
+ nexthdr = ipha->ipha_protocol;
+ src_val = (uint32_t)ipha->ipha_src;
+ /*
+ * Catch IPv4 fragment case here. IPv6 has nexthdr == FRAG
+ * for its equivalent case.
+ */
+ if (mac_src_ipv4_fanout ||
+ (ntohs(ipha->ipha_fragment_offset_and_flags) &
+ (IPH_MF | IPH_OFFSET)) != 0) {
+ goto src_based_fanout;
}
+ }
+ if (remlen < MIN_EHDR_LEN)
+ return (-1);
+ whereptr = (uint8_t *)ip6h + hdr_len;
+
+ /* If the transport is one of below, we do port/SPI based fanout */
+ switch (nexthdr) {
+ case IPPROTO_TCP:
+ case IPPROTO_UDP:
+ case IPPROTO_SCTP:
+ case IPPROTO_ESP:
+ /*
+ * If the ports or SPI in the transport header is not part of
+ * the mblk, do src_based_fanout, instead of calling
+ * pullupmsg().
+ */
+ if (mp->b_cont == NULL || whereptr + PORTS_SIZE <= mp->b_wptr)
+ break; /* out of switch... */
+ /* FALLTHRU */
+ default:
+ goto src_based_fanout;
+ }
- /* Do src based fanout if below tunable is set to B_TRUE. */
- if (mac_src_ipv4_fanout)
- goto ipv4_src_based_fanout;
-
- /* If the transport is TCP, we try to do port based fanout */
- if (ip4h->ip_p == IPPROTO_TCP) {
- int hdr_len;
-
- hdr_len = ip4h->ip_hl << 2;
- /* set whereptr to point to tcphdr */
- whereptr = (uint8_t *)ip4h + hdr_len;
-
- /*
- * If ip4h does not hold the complete ip header
- * including options, or if both ports in the TCP
- * header are not part of the mblk, do src_based_fanout
- * (the second case covers the first one so we only
- * need one test).
- */
- if (mp->b_cont != NULL &&
- whereptr + PORTS_SIZE > mp->b_wptr)
- goto ipv4_src_based_fanout;
-
- hash = HASH_ADDR(ip4h->ip_src.s_addr,
- *(uint32_t *)whereptr);
+ switch (nexthdr) {
+ case IPPROTO_TCP:
+ hash = HASH_ADDR(src_val, *(uint32_t *)whereptr);
+ *indx = COMPUTE_INDEX(hash, mac_srs->srs_tcp_ring_count);
+ *type = OTH;
+ break;
+ case IPPROTO_UDP:
+ case IPPROTO_SCTP:
+ case IPPROTO_ESP:
+ if (mac_fanout_type == MAC_FANOUT_DEFAULT) {
+ hash = HASH_ADDR(src_val, *(uint32_t *)whereptr);
*indx = COMPUTE_INDEX(hash,
- mac_srs->srs_tcp_ring_count);
- *type = OTH;
+ mac_srs->srs_udp_ring_count);
} else {
- /* For all other protocols, do source based fanout */
- goto ipv4_src_based_fanout;
+ *indx = mac_srs->srs_ind % mac_srs->srs_udp_ring_count;
+ mac_srs->srs_ind++;
}
- } else {
- *indx = 0;
*type = OTH;
+ break;
}
return (0);
-ipv6_src_based_fanout:
- hash = HASH_ADDR(V4_PART_OF_V6(ip6h->ip6_src), (uint32_t)0);
- *indx = COMPUTE_INDEX(hash, mac_srs->srs_oth_ring_count);
- *type = OTH;
- return (0);
-
-ipv4_src_based_fanout:
- hash = HASH_ADDR(ip4h->ip_src.s_addr, (uint32_t)0);
+src_based_fanout:
+ hash = HASH_ADDR(src_val, (uint32_t)0);
*indx = COMPUTE_INDEX(hash, mac_srs->srs_oth_ring_count);
*type = OTH;
return (0);
diff --git a/usr/src/uts/common/sys/fcntl.h b/usr/src/uts/common/sys/fcntl.h
index a0c4191b1c..f3ca84fff1 100644
--- a/usr/src/uts/common/sys/fcntl.h
+++ b/usr/src/uts/common/sys/fcntl.h
@@ -36,6 +36,8 @@
* contributors.
*/
+/* Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved. */
+
#ifndef _SYS_FCNTL_H
#define _SYS_FCNTL_H
@@ -82,6 +84,7 @@ extern "C" {
#define O_XATTR 0x4000 /* extended attribute */
#define O_NOFOLLOW 0x20000 /* don't follow symlinks */
#define O_NOLINKS 0x40000 /* don't allow multiple hard links */
+#define O_CLOEXEC 0x800000 /* set the close-on-exec flag */
/*
* fcntl(2) requests
diff --git a/usr/src/uts/common/sys/file.h b/usr/src/uts/common/sys/file.h
index f402bc7079..03acc088c2 100644
--- a/usr/src/uts/common/sys/file.h
+++ b/usr/src/uts/common/sys/file.h
@@ -26,6 +26,8 @@
/* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
/* All Rights Reserved */
+/* Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved. */
+
#ifndef _SYS_FILE_H
#define _SYS_FILE_H
@@ -113,6 +115,8 @@ typedef struct fpollinfo {
#define FSEARCH 0x200000 /* O_SEARCH = 0x200000 */
#define FEXEC 0x400000 /* O_EXEC = 0x400000 */
+#define FCLOEXEC 0x800000 /* O_CLOEXEC = 0x800000 */
+
#ifdef _KERNEL
/*
diff --git a/usr/src/uts/common/syscall/open.c b/usr/src/uts/common/syscall/open.c
index 6559900158..edb04c824b 100644
--- a/usr/src/uts/common/syscall/open.c
+++ b/usr/src/uts/common/syscall/open.c
@@ -27,6 +27,9 @@
/* All Rights Reserved */
/*
+ * Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved.
+ */
+/*
* Portions of this source code were derived from Berkeley 4.3 BSD
* under license from the Regents of the University of California.
*/
@@ -226,6 +229,9 @@ noxattr:
* falloc reserved.
*/
setf(fd, fp);
+ if ((filemode & FCLOEXEC) != 0) {
+ f_setfd(fd, FD_CLOEXEC);
+ }
return (fd);
} else {
/*
@@ -252,6 +258,9 @@ noxattr:
fp->f_count++;
mutex_exit(&fp->f_tlock);
setf(fd, fp);
+ if ((filemode & FCLOEXEC) != 0) {
+ f_setfd(fd, FD_CLOEXEC);
+ }
releasef(dupfd);
}
return (fd);
generated by cgit v1.2.3 (git 2.39.1) at 2024-11-23 13:22:48 +0000